FPGA-based · full-precision edge AI
Empowering Autonomous Intelligence at the Edge
Invotet modules don’t ride a repurposed mobile GPU — and they aren’t a chip-down ASIC waiting on tape-out. The Invotet Unified Engine runs on an FPGA fabric you can buy today, derived from the full attention kernel — matrix multiplication, quantization, normalization, and data movement — and optimized for the operations that actually run a modern model.
TerraBot X
BF16 inference engine · PCIe interface · 8.4 GOPS/W power efficiency
Supports BF16, INT4, FP4 · Flash attention · Tensor parallelism · 8,192-entry hardware trace buffer
Efficiency
20×
vs Jetson · per watt
Utilization
95%
sustained transformer utilization
Range
−40 / +85
°C mil-spec operating
Telemetry
8,192
cycle-accurate trace timestamps
Qualified to
Preliminary · datasheet on request
MIL-STD-810H
Method 514.8 vibration
MIL-STD-810H
Method 516.8 shock
−40 to +85 °C
Mil-spec operating range
IP67
Dust + immersion ingress · boxed
30,000 ft
Altitude qualified · AeroScale V1
Secure boot
Hardware root of trust · per-module attestation
Designed for
The buyers who can't ship on commercial silicon
- Vertical detail
Aerial robotics & drones
Onboard inference under battery budget
- Vertical detail
Ground robotics
SLAM, sensor fusion · MIL-STD-810H
Defense ISR
Sensor fusion at the tactical edge
One FPGA engine · two qualifications
Same compute. Qualified for the environment you fly or drive across.
The FPGA fabric and Unified Engine are identical across the line. The SKUs differ by what each module is hardened and tested against — altitude and vibration for aerial, shock and ingress for ground robotics.
FPGA-based · aerial-grade
AeroScale V1
FPGA-based, sub-10W AI accelerator for autonomous drones, delivery platforms, and high-altitude aerial systems.
- Operating range
- −40 to +85 °C
- Qualified for
- Aerial-grade · MIL-STD-810H vibration · 30,000 ft altitude
FPGA-based · robotics-grade
TerraBot X
FPGA-based edge AI accelerator for ground robotics — real-time SLAM, multi-sensor fusion, and on-platform reasoning under mil-spec shock and vibration.
- Operating range
- −40 to +85 °C
- Qualified for
- Robotics-grade · MIL-STD-810H shock + vibration · IP67 boxed
Why the module
Up to 20× more efficient — and training-equivalent accurate at inference.
Up to 20× efficiency
A unified compute engine — systolic and vector processing in one — purpose-built for transformer workloads. Smallest logic footprint, highest utilization, up to 20× more efficient than NVIDIA Jetson.
Sustainable autonomy
Frontier-class models inside a sub-15W envelope. AI fits inside the battery or solar budget — Size, Weight, and Power optimized for every module.
Transformer-grade fidelity
BF16-native execution preserves training-equivalent accuracy at 95% sustained utilization, with native flash attention and hardware tensor-parallel sync. A cycle-accurate hardware trace buffer and compile-graph-to-hardware specialization make every inference verifiable and tuned to the workload.
Mil-spec environment
Operate from −40 °C to +85 °C. Survive MIL-STD-810H shock, vibration, altitude, and IP67 ingress — the regimes that disqualify commercial silicon.
GPT-native logic
Matrix multiplication, softmax, element-wise operations, and the rest of the transformer operator set run natively in purpose-built logic — no general-purpose emulation tax.
Secure by design
Hardware root of trust, signed firmware, and per-module attestation keep both the model and the device tamper-evident.
Invotet SDK
Compile once. Deploy to every Invotet module.
A unified Python SDK that ingests PyTorch, ONNX, and HuggingFace checkpoints, quantizes for Invotet modules, and ships a deterministic runtime to the device. No CUDA in the loop.
Framework
PyTorch
Trace or torch.export checkpoints compile directly with no rewrite.
Framework
ONNX
Standards-based interchange — compile any ONNX-exported model.
Framework
HuggingFace
transformers checkpoints land on Invotet through a one-line loader.
For developers
Real entry points, not a wall of marketing
- Open
Documentation
API reference, runtime, and module integration guides
- Open
Quickstart
From a HuggingFace checkpoint to a deployed module in minutes
- Open
Model Explorer
Tested checkpoints across LLM, VLM, and perception families
- Coming soon
GitHub
Open-source examples, model recipes, and SDK source
Need datasheets, qualification reports, or NDA?
Most procurement conversations start with the datasheet plus the relevant qualification report. Open a sales conversation and we will route both to your inbox the same day.